Encapsulated and Anonymized Network Video Traffic Classification With Generative Models

In recent years, anonymization and encrypted web applications have become increasingly popular among the public. Although the Internet communication process already has complete privacy protection, network security technologies continue to evolve, and technologies for exploring the commercial behavioral characteristics of network users' fingerprints are also accelerating. In this paper, we picked benchmark public datasets, including anonymously viewing video streaming traffic using The Onion Router (Tor) network technology, establishing tunnel encapsulated video streaming packets using Virtual Private Network (VPN) technology, and viewing video streaming traffic scenarios in normal mode. We combined video streaming application types and names to construct a new dataset for multi-class classification problem. Unlike the frequently used classification discrimination models, we used generative models, including Restricted Boltzmann Machines (RBM) and Deep Belief Networks (DBN), having an advantage of features extraction learning to construct models for probabilistic judgments, and better performance, that has been verified and supported in several works. The experimental results show the classification accuracy is as high as 0.94 for eight categories and 0.97 for three categories. Its overall classification performance is good, and different classification performance metrics are balanced.